|Cook, A. Z.|
Submitted to: Phytopathology
Publication Type: Abstract Only
Publication Acceptance Date: 4/1/2008
Publication Date: 7/26/2008
Citation: Cook, A., Bock, C.H., Parker, P.E., Gottwald, T.R. 2008. Automating the assessment of citrus canker symptoms with image analysis. Phytopathology. 98:S41 Interpretive Summary:
Technical Abstract: Citrus canker (CC, caused by <i>Xanthomonas citri</i>) is a serious disease of citrus in Florida and other citrus-growing regions. Severity of symptoms can be estimated by visual rating, but there is inter- and intra-rater variation. Automated image analysis (IA) may offer a way of reducing some of the errors in accuracy and precision. Actual disease was measured on individual leaves in 4 data sets of digital leaf images (with 65, 22, 123 and 200 images respectively, actual severity range 0-59%). The leaf images were assessed by three raters (VRs) and by automation with IA (Assess, APS, St Paul, MN). The groups of leaves varied in image quality (leaf quality, lighting, reflection, focus and background uniformity). Group 1 and 4 had the best quality and were most consistent. Group 2 and 3 were poorer quality. Regression analysis showed that group 1 and 4 were assessed with greatest accuracy and precision by VRs (<i>r2</i>=0.86-0.94 and 0.78-0.84, respectively). VRs performed less well on group 2 and 3 (<i>r2</i>=0.57-81 and 0.65-0.77, respectively). Automated IA of groups 1 (<i>r2</i>=0.87) and 4 (<i>r2</i>=0.82) was more accurate and precise compared to group 2 (<i>r2</i>=0.33) and 3 (<i>r2</i>=0.66). Whether assessed by VRs or measured by IA the data were heteroscedastic with respect to actual severity. Automation of IA can provide estimates of severity within the range assessed by VRs - provided image quality is good and uniform. Advantages of IA can include time saving and reduced error.